Learned volatility forecasts
Forecasts how the whole volatility surface is likely to move.
Deep-learning models trained on historical options market data, forecasting across strikes and expiries rather than a single volatility number.
Incorporated in Hong Kong · 2024
Machine learning for options portfolio risk.
Necessity Computing licenses a hosted API that uses machine-learning forecasts of the volatility surface to work out which adjustment best serves an options portfolio’s objective while keeping it within set limits. Clients keep their own accounts and make their own trading decisions.
Portfolio in, risks out.
The client’s system sends data to the API and receives risks back. Decisions and execution stay with the client.
- Step 01 client system
Client sends
The client sets an objective and limits on delta, gamma, vega and theta. Its system sends the portfolio whenever the client wants risks recalculated, up to real time.
- Step 02 necessity engine
Engine forecasts
Machine-learning models forecast how the volatility surface is likely to move. Using that forecast, the engine works out the adjustment that best serves the client’s objective while keeping the portfolio within its limits, including under stressed conditions.
- Step 03 api → client
API returns
The API returns a proposed adjustment, or no change if no adjustment is worth its trading cost. The client decides whether to act and executes in its own accounts.
current vs forecast surface
currentML forecast
Where the work lives.
The software has three parts: machine-learning volatility models, the engine that weighs adjustments against the client’s objective and limits, and the API that delivers results to the client’s systems.
Cost-aware adjustments
Works out which change, if any, best serves the client’s objective within its limits.
A change is proposed only when its expected benefit outweighs its trading cost, and never if it would breach the client’s limits, including under stressed conditions.
Hosted API
Licensed on subscription to institutions managing options exposure.
Runs on our servers and returns results to the client’s own systems. Nothing to install; no UI required.
01 / Technological core
Built at the seam between quant and engineering.
We operate at the intersection of quantitative finance, machine learning and software engineering: models, data pipelines and delivery built as one system. A model is not a notebook; it is a pipeline.
02 / Look ahead
Look ahead, don’t react.
Exchanges and risk tools report where a portfolio’s exposure is now. Our software acts on a forecast of where volatility is heading, and proposes nothing when no change is worth its cost.
Statistical validation over qualitative judgment.
Models are trained on historical options market data and judged out of sample on one question: how well the software serves the client’s objective while keeping the portfolio within its limits as conditions change. That validation is in progress.
- i.Forecasts are learned from market datanot assumed from a fixed formula
- ii.Objectives and limits are set by the clientnot by us
- iii.Changes must earn their costno adjustment unless its expected benefit exceeds its trading cost
- iv.Evaluation is out of sampleon data the models were not trained on